An exploratory study of air emissions associated with shale gas development and production in the Barnett Shale

Abstract

Information regarding air emissions from shale gas extraction and production is critically important given production is occurring in highly urbanized areas across the United States. Objectives of this exploratory study were to collect ambient air samples in residential areas within 61 m (200 feet) of shale gas extraction/production and determine whether a “fingerprint” of chemicals can be associated with shale gas activity. Statistical analyses correlating fingerprint chemicals with methane, equipment, and processes of extraction/production were performed. Ambient air sampling in residential areas of shale gas extraction and production was conducted at six counties in the Dallas/Fort Worth (DFW) Metroplex from 2008 to 2010. The 39 locations tested were identified by clients that requested monitoring. Seven sites were sampled on 2 days (typically months later in another season), and two sites were sampled on 3 days, resulting in 50 sets of monitoring data. Twenty-four-hour passive samples were collected using summa canisters. Gas chromatography/mass spectrometer analysis was used to identify organic compounds present. Methane was present in concentrations above laboratory detection limits in 49 out of 50 sampling data sets. Most of the areas investigated had atmospheric methane concentrations considerably higher than reported urban background concentrations (1.8–2.0 ppm_v). Other chemical constituents were found to be correlated with presence of methane. A principal components analysis (PCA) identified multivariate patterns of concentrations that potentially constitute signatures of emissions from different phases of operation at natural gas sites. The first factor identified through the PCA proved most informative. Extreme negative values were strongly and statistically associated with the presence of compressors at sample sites. The seven chemicals strongly associated with this factor (o-xylene, ethylbenzene, 1,2,4-trimethylbenzene, m- and p-xylene, 1,3,5-trimethylbenzene, toluene, and benzene) thus constitute a potential fingerprint of emissions associated with compression.

Implications: Information regarding air emissions from shale gas development and production is critically important given production is now occurring in highly urbanized areas across the United States. Methane, the primary shale gas constituent, contributes substantially to climate change; other natural gas constituents are known to have adverse health effects. This study goes beyond previous Barnett Shale field studies by encompassing a wider variety of production equipment (wells, tanks, compressors, and separators) and a wider geographical region. The principal components analysis, unique to this study, provides valuable information regarding the ability to anticipate associated shale gas chemical constituents.

Introduction

With advances in horizontal drilling and hydraulic fracturing techniques, production of natural gas from hydrocarbon-rich shale formations, or shale gas, is bringing drilling and production operations to regions of the United States that have seen little or no similar activity in the past (Lev-On and Levy, 2012). Over the last several years, natural gas drilling and production have become commonplace in several U.S. shale formations: the Marcellus Shale in Pennsylvania; the Barnett and Eagle Ford shales in Texas; and the Niobrara Shale in Colorado. In particular, exploration and production of unconventional natural gas in the Texas Barnett Shale increased dramatically in the last decade, with the number of shale gas wells increasing from 726 in 2000 to 14,886 in 2010 (Texas Railroad Commission, 2012). Shale gas extraction and production in the Texas Barnett Shale has generated considerable interest in potential environmental impacts.

Shale gas extraction and processing involves numerous processes, including well development (pad preparation, well drilling, and well completion, which includes hydraulic fracturing and flowback); gas production; gas processing (separation, dehydration, desulfurization); condensate storage; gas compression; and gas transmission (Armendariz, 2009; McKenzie et al., 2012). In terms of air pollutants, potential compounds of concern associated with these processes include

•	criteria air pollutants (particulate matter, nitrogen oxides, sulfur oxides, and carbon monoxide), due to compressor engines (Armendariz, 2009; Bar-Ilan et al., 2008; Eastern Research Group, 2011);
•	volatile organic compounds, many of which are precursors of ground-level ozone formation, as well as hazardous air pollutants, including benzene and formaldehyde (Boyer, 2010; Eastern Research Group, 2011; Hendler et al., 2009; Olaguer, 2012; Pring et al., 2010; Safitri et al., 2011; XTO Energy, 2010);
•	methane, a greenhouse gas (Dedikov et al., 1999; Eastern Research Group, 2011; U.S. Environmental Protection Agency [EPA], 1996, 2006; Hendler et al., 2009; Howarth et al., 2011; Pring et al., 2010; Safritri et al., 2011; XTO Energy, 2010); and
•	odor-causing compounds, such as hydrogen sulfide and other reduced sulfides (Coward and Barron, 1983; Dawodu and Meisen, 1989; Eapi et al., 2013; Mazumdar et al., 1974; Kunkel, 1977; Tuan et al., 1995; Yuhua et al., 2006).

Previous studies of natural gas development and production on air quality

With the recent increase in natural gas production volume, more studies are examining the impact of natural gas operations on air quality. Katzenstein et al. (2003) conducted two intensive surface air discrete sampling studies over the Anakarko Fossil Fuel Basin in the southwestern United States, and found substantial regional atmospheric methane and nonmethane hydrocarbon pollution over parts of Texas, Oklahoma, and Kansas attributed to the oil and gas industry. The study suggested that total U.S. natural gas emissions may have been underestimated. This study was conducted prior to production volume in the Barnett Shale increasing substantially. Schnell et al. (2009) observed high wintertime ozone levels near the Jonah-Pinedale Anticline natural gas field in western Wyoming. Petron et al. (2012) analyzed daily air samples collected at a tall-tower monitoring site, as well as on-road survey data, for the Denver-Julesburg Fossil Fuel Basin in the Colorado Northern Front Range, which contains over 20,000 active natural gas and condensate wells. They found that a mix of natural gas venting emissions and flashing emissions from condensate tanks can explain observed alkane ratios. McKenzie et al. (2012) found that residents living <½ mile from shale gas wells in Garfield County, Colorado, were at greater risk for health impacts, due to subchronic exposure to trimethylbenzenes and xylenes, and cumulative cancer risk due to benzene. Stephenson, Valle, and Riera-Palou (2011) compared greenhouse gas emissions from conventional natural gas and shale gas production. They estimated that shale gas typically has a well-to-wheel emission intensity 1.8–2.4% higher than conventional gas, arising mainly from higher methane releases during well completion. Burnham and Clark (2012) compared life-cycle greenhouse emissions from use of shale gas and conventional natural gas. Their base case found that shale gas life-cycle emissions are 6% lower than conventional natural gas. However, the range in values for shale and conventional gas overlapped, so there was statistical uncertainty whether shale gas emissions were indeed lower than conventional gas.

Within Texas, Burklin and Heaney (2006) of Eastern Research Group conducted a field survey of natural gas compressor engine sizes and types in the eastern part of Texas for the Houston Advanced Research Center. They found that 50–73% of the well-head engine capacity in East Texas is composed of engines greater than 500 hp, depending on the region. In a project conducted for the Texas Environmental Research Consortium, Hendler of URS Corporation and colleagues (2009) measured emissions from oil and condensate storage tanks in East Texas by directly monitoring the flow rates of gases escaping from storage tank vents and sampling the vent gases for chemical composition. The study included measurements from 11 condensate tank batteries in Denton County and 2 in Parker County in North Texas.

Armendariz (2009) estimated emissions of volatile organic compounds (VOCs), nitrogen oxides, hazardous air pollutants (HAPs), methane, carbon dioxide, and nitrous oxide from natural gas drilling and production in the Barnett Shale. His study assimilated information from a variety of previous studies (EPA AP-42 emission factors, the URS study, and a Gas Research Institute/EPA study) in order to estimate emissions from compressor engines; condensate tanks; well drilling, hydraulic fracturing pump engines, and well completions; and production, process, and transmission fugitives. Among the natural gas sources, compressor engines and condensate tanks were found to be the largest sources of smog-forming compounds and hazardous air pollutants. Compressor engines and fugitive emissions from all source types were found to be the largest sources of greenhouse gas emissions.

The Fort Worth Natural Gas Air Quality Study (2011), conducted by Eastern Research Group under contract to the City of Fort Worth, surveyed 388 sites including 375 well pads, eight compressor stations, one gas processing plant, a saltwater treatment facility, a drilling operation, a fracking operation, and a completion operation. Repeat visits were conducted at two sites. At these 388 sites, measurements were conducted using an infrared (IR) camera (detects large emission sources with concentrations of methane, ethane, propane, and butane >10,000 ppm) and a toxic vapor analyzer that measures hydrocarbons with concentrations as low as 0.5 ppm. If the IR camera identified high levels of emissions, then a high-flow sampler was used to capture gas emitting from a component. At 164 locations, Summa Passivated Stainless Steel Canisters were used to collect gas samples from selected emission points for methane, VOC, and HAP analysis. The study found that low-toxicity pollutants (e.g., methane, ethane, propane and butane) accounted for approximately 98% of the citywide emissions. However, several pollutants with relatively high toxicities (e.g., benzene) were also emitted, although in considerably lower quantities. For the relatively few sites with multiple large compressor engines, the modeling analysis found some locations beyond the city's required 600-foot setback distance to have estimated acrolein and formaldehyde concentrations greater than protective health-based screening levels published by the Texas Commission on Environmental Quality.

This study encompasses a wider variety of production equipment (wells, tanks, compressors, and separators) than several of the previous studies, and a wider geographical region (seven counties) compared with the Fort Worth study. The principal components analysis (PCA) is unique to this study and provides preliminary information regarding the ability to anticipate associated chemical constituents.

Study objectives

The objectives of this exploratory study were to answer the following questions:

1.	What chemicals are present at what concentrations in ambient air samples in residential areas in the DFW Metroplex near shale gas wells?
2.	Are concentrations of chemicals associated with methane, the primary product of shale gas production?
3.	Is there a relationship among the various chemicals present in residential ambient air samples near shale gas wells?
4.	Do patterns of correlation among particular compounds provide signatures of emissions associated with different technological equipment or processes of shale gas well extraction and production operations?

Experimental Methods

Field sampling

Ambient air monitoring in residential areas near natural gas production facilities was performed using certified sterilized evacuated stainless steel 6-L Summa canisters with 24-hr flow regulators (certified mass-flow 24-hr meter). The flow valves were regulated to allow for continuous sampling over a 24-hr period.

Fifty ambient air sampling events occurred in residential areas where shale gas extraction and production was occurring in six counties in the Dallas/Fort Worth (DFW) Metroplex from 2008 to 2010. In all, 39 separate locations were sampled, resulting in 50 sets of monitoring data. Nine locations were sampled more than once during the time period, with seven locations monitored twice and two were monitored three times during the time period. Typically, multiple sampling events at the same location were performed at different times of the year and therefore allowed for seasonal comparison of air composition. The locations were identified by clients that requested monitoring. Twenty-four-hour passive samples were collected using summa canisters.

Figure 1 identifies general location of sampling sites in DFW Metroplex. Of the 39 locations, 20 locations were identified as being in low-density residential areas (5–10-acre lots) with minimal potential for any other confounding source emissions other than light transportation emissions; 8 monitoring locations were in higher-density residential areas with little, if any, potential for confounding source emissions other than light transportation emissions; and 2 of the monitoring locations were in residential areas with potential confounding sources present. One of the locations was in the general location of a landfill, which was possibly a contributing factor for presence of methane; however, this particular location was the single instance of no methane detected. The other location was in the general vicinity of an airport (within 2 miles). To quantity potential methane emissions from an airport, fuel type, aircraft type, engine type, engine load and altitude of aircraft, and air traffic must be known. Detailing of emission sources to this degree was outside the scope of this paper.

Figure 1. Sampled locations in DFW Metroplex.

Sampling procedures followed American Society for Testing and Materials (ASTM) Method D-1357-95 (2011), Standard Practice for Planning the Sampling of the Ambient Atmosphere. Canister locations were verified with global positioning system (GPS) coordinates. A certified laboratory (GD Air, Plano, TX) conducted three analytical tests on the canister samples using a Hewlett-Packard (Agilent; Santa Clara, CA) gas chromatograph/mass spectrometer: EPA Compendium Method Toxic Organics (TO-14A), Light Hydrocarbons, and Tentatively Identified Compounds (TICs). The TO-14A test is performed to capture a broad range of volatile organic compounds that have been determined to be stable when stored in pressurized canisters; however, it is not effective for lower-chain hydrocarbons C1–C6. The Light Hydrocarbon test is able to detect lower-chain hydrocarbons (C1–C6). This would include compounds of ethane, methane, propane, and butane, which are smaller hydrocarbons. Identification of Tentatively Identified Compounds allows reporting of compounds that the instrumentation can detect but that the analysis is not targeting specifically.

Equipment at natural gas production sites can vary widely. In order to evaluate potential emission sources, equipment at pad sites nearest each residential sampling site and also present in the general area was inventoried at the time of monitoring or shortly thereafter. To determine the proximity of the equipment to each sampling location, three circular zones were established at radii of 61 m (200 feet), 610 m (2000 feet), and 1.6 km (1 mile). Equipment was inventoried within the radii, and any potential major contributing source related to natural gas was noted. Equipment inventoried included the number of wells, tanks, compressors, and separators present within each radius. Table 1summarizes the equipment found within each radius for each sampling location.

Table 1. Equipment and zones of influence

			Equipment within 61 m of Sampling Site				Equipment within 610 m of Sampling Site				Equipment within 1.6 km of Sampling Site
County	Site	Times Sampled	No. of Wells	No. of Tanks	No. of Comp.	No. of Sep.	No. of Wells	No. of Tanks	No. of Comp.	No. of Sep.	No. of Wells	No. of Tanks	No. of Comp.	No. of Sep.
Parker	A	1	0	0	0	0	4	6	9	1	14	18	18	3
	B	1	0	0	0	0	4	6	9	1	14	18	18	3
	C	1	0	0	0	0	0	0	9	1	10	12	18	3
	D	2	0	0	0	0	0	0	0	0	8	6	2	1
	E	1	0	0	0	0	0	0	0	0	8	6	2	1
	F	1	0	0	0	0	0	0	0	0	8	6	2	1
	G	1	0	0	0	0	0	0	0	0	2	4	1	2
	H	1	1	0	0	0	1	0	0	0	1	0	0	0
		9
Tarrant	A	2	0	0	0	0	0	0	0	0	2	2	0	0
	B	2	0	0	0	0	0	0	0	0	2	2	0	0
	C	2	0	0	0	0	0	0	0	0	2	2	0	0
	D	1	1	0	0	0	5	10	3	2	21	40	11	8
	E	1	1	0	0	0	1	0	0	0	3	4	1	1
	F	1	1	0	0	0	1	0	0	0	2	2	1	1
	G	1	1	0	0	0	4	2	1	1	4	4	2	2
		10
Johnson	A	1	0	0	4	2	1	4	4	2	1	4	4	2
	B	1	0	0	4	2	1	4	4	2	1	4	4	2
	C	1	0	0	0	0	1	2	0	0	8	14	4	2
	D	2	0	0	0	0	1	2	0	0	8	14	4	2
	E	1	2	12	2	2	2	12	2	2	5	12	2	2
	F	1	1	2	0	0	1	2	0	0	1	2	0	0
		7
Denton	A	1	0	0	0	0	4	4	9	6	4	4	9	6
	B	1	0	0	0	0	4	4	9	9	4	4	9	6
	C	2	0	0	0	0	0	0	0	0	3	4	9	6
	D	3	0	0	0	0	2	2	6	1	4	4	9	6
	E	1	0	0	0	0	2	2	6	1	4	4	9	6
	F	1	0	0	0	0	2	2	4	6	4	4	9	6
	G	1	0	1	1	1	2	4	5	6	4	4	9	6
	H	1	0	0	0	0	2	0	0	0	2	0	0	0
	I	1	0	0	0	0	2	0	0	0	2	0	0	0
	J	1	0	0	0	0	0	0	0	0	5	6	4	2
	K	1	4	4	1	1	6	6	1	1	24	24	4	6
	L	2	0	0	0	0	2	0	0	0	6	6	0	1
	M	3	0	0	0	0	2	2	0	0	22	14	5	8
	N	1	2	2	0	0	6	0	1	1	12	12	12	4
		20
Somer-ville	A	1	0	0	0	0	0	0	0	0	4	6	2	0
		1
Wise	A	1	2	4	0	0	2	4	0	0	4	8	2	0
	B	1	0	0	0	0	0	0	0	0	4	8	2	0
	C	1	2	4	0	0	2	4	0	0	4	8	2	0
		3
Notes: Comp. = compressors; Sep. = separators.

Meteorological conditions on the dates of testing were retrieved from the National Climatic Data Center Quality Controlled Local Climatological Data Web site (www.ncdc.noaa.gov), using the airport closest to each sampled locations. Ambient air monitoring occurred throughout the year and thus included seasonal variations. Temperatures varied from 26 to 104 °F, and wind speeds varied from 0 to 26 mph. Although variations in wind speed and atmospheric turbulence would have impacted the magnitude of concentrations measured, these variations would not have impacted the ratios of compounds measured, because molecules of different kinds would have been diluted and dispersed similarly. For example, if the concentration of benzene was twice that of toluene at a given location on a day with high wind speeds and high atmospheric turbulence, the concentration of benzene would still be twice that of toluene at that same location on a day with low wind speeds and low levels of atmospheric turbulence, assuming that the source was still emitting benzene at double the rate of toluene. The magnitude of the concentrations would differ on the days with high and low wind speeds and turbulence, but the ratio of the benzene to toluene concentration would still be 2. This means that the correlation relationships found between chemicals in this study would have been the same even if the meteorology on the days sampled had been different.

Statistical analysis

As discussed above, there are few data sets documenting the potential air emissions of shale gas development and production. Given the logistical difficulties of obtaining representative samples, this paper presents an exploratory analysis that raises hypotheses to be explored in further work. In keeping with this exploratory approach, a well-understood analytical approach was applied to generate hypotheses, in part because alternative approaches mostly require larger data sets to be effective.

Summary statistics were calculated to determine basic information related to chemical constituents. The minimum value, maximum value, median value, mean value, and standard deviation for each chemical were determined. The minimum value for each compound is its gas chromatograph detection limit.

Further statistical analysis was performed using a Pearson's product-moment correlation coefficient (Pearson's correlation coefficient or Pearson's r) to determine the strength of relationship between the concentration of methane and the concentration of other volatile organic compounds. For this analysis, all nondetect observations were coded to the detection limit. The concentration data were not normally distributed for many chemicals, so all concentrations were transformed to natural logarithms prior to calculating correlations, and prior to other analyses.

To provide a more in-depth analysis of the relationships of chemicals to methane and chemicals to each other, a principal components analysis (PCA) was performed using a statistical software package. With PCA, data are transformed from a large set of related variables to a smaller set of uncorrelated variables. The newly created variables are called principal components (PCs), or factors. These factors are interpreted as indicating latent, unmeasured variables that explain linearly correlated patterns in the original variables. As such, factors can then be analyzed in relation to other measured variables that are potential predictors of variation in the original variables. The PCA was performed on data that were edited to contain only chemicals measured above detection limits in at least 10% of samples (i.e., at least five detectable measurements). The average number of detected values in the edited set of 36 chemicals was 19, and the median was 14. The PCA was performed on the correlation matrix of the selected data.

An analysis of variance (ANOVA) F test was then performed to test the null hypothesis that sample site scores for PCA factor 1 and factor 2 are independent of particular phases and equipment associated with nearby gas well operations.

Results and Discussion

Air sampling confirmed the presence of methane and 101 other chemicals in the atmosphere in and around sampled residential sites in the DFW Metroplex where unconventional shale gas extraction and production was the predominant emission activity. Table 2 provides summary statistics for the chemical compounds measured. Approximately 20 of the 101 (20%) chemicals identified are listed as HAPs according to the EPA, including benzene, 1,3-butadiene, carbon disulfide, carbonyl sulfide, chloromethane, tetrachloroethane, toluene, and xylene.

Table 2. Summary statistics

CAS No.	Chemical	Min (ppbv)	Max (ppbv)	Median (ppbv)	Mean (ppbv)	SD	No. of ND
74828	Methane (ppmv)	1.9	457	2.7	11.99	63.58	1
71432	Benzene	0.6	592	0.89	18.53	83.75	11
67663	Chloroform	0.2	2.58	0.3	0.45	0.46	46
74873	Chloromethane/Methyl chloride	0.25	5.33	0.6	0.68	0.71	17
75718	Dichlorodifluoromethane (F12)	0.25	1.13	0.45	0.48	0.17	9
76142	Dichlorotetrafluoroethane (F114)	0.2	1	0.3	0.36	0.17	47
107062	1,2-Dichloroethane (EDC)	0.2	1	0.3	0.34	0.15	49
75092	Dichloromethane/Methylene chloride	0.2	1	0.3	0.34	0.15	49
100414	Ethylbenzene	0.2	113	0.53	4.42	16.03	23
87683	Hexachlorobutadiene	0.25	2.6	0.7	0.73	0.46	42
100425	Styrene	0.2	43.4	0.37	1.91	6.22	26
79345	1,1,2,2-Tethrachloroethane	0.2	2.06	0.3	0.37	0.28	49
127184	Tetrachloroethene (PCE)	0.2	2.43	0.3	0.33	0.39	47
108883	Toluene/Methylbenzene	0.34	276	2.55	19.45	48.77	3
108678	1,3,5-Trimethylbenzene	0.2	9.95	0.59	1.43	2.12	25
95636	1,2,4-Trimethylbenzene	0.2	60.4	0.4	3.45	10.79	27
120821	1,2,4-Trichlorobenzene	0.28	13.5	0.67	1.12	1.93	10
79016	Trichloroethene (TCE)	0.2	60.9	0.3	1.58	8.48	47
75694	Trichlorofluoromethane (F11)	0.2	1	0.3	0.34	0.15	45
1330207	m- and p-Xylene	0.25	221	1.68	15.69	43.1	7
95476	o-Xylene	0.2	39.4	0.85	3.19	6.7	15
75150	Carbon disulfide	0.7	103	4	11.75	20.5	22
463581	Carbonyl sulfide	0.3	36.7	1.41	4.22	7.1	40
624920	Dimethyl disulfide	0.3	200	1.93	15	31.56	29
20333395	Methyl ethyl disulphide	0.3	145	1.78	11.18	24.27	31
611143	Ethylmethylbenzene	0.3	42.8	1.4	3.15	6.74	47
2179604	Methyl propyl disulfide	0.3	41.6	1.4	2.59	5.71	49
110816	Diethyl disulfide	0.3	32.7	1.5	3.15	5.92	43
53966362	Ethyl, methylethyl disulfide	0.3	46.7	1.4	3.68	8.87	46
3658808	Dimethyl trisulfide	1.2	46.3	1.52	8.02	14.86	37
30453317	Ethyl n-propyl disulfide	0.3	25.2	1.4	2.25	3.48	48
95636	Trimethylbenzene	0.3	366	1.4	15.18	58.39	46
11020214	Undecane	0.3	72	1.4	3.05	9.88	49
2082613	1-Methyl propenylbenzene	0.3	51	1.4	2.63	6.96	49
112403	Dodecane	0.3	29	1.4	2.19	3.91	49
767599	1-methyl-1H Indene	0.3	79	1.4	3.19	10.86	49
768490	2-Methyl propenyl benzene	0.3	95.9	1.4	3.53	13.22	49
103651	Propylbenzene	0.3	23.5	1.4	2.08	3.16	49
25340174	Diethylbenzene	0.3	93.4	1.4	4.14	13.56	48
19876	Methyl-methylethylbenzene/methylcumene	0.3	84.7	1.4	3.31	11.65	48
110189	Tetramethylbenzene	0.3	36.4	1.4	2.76	5.69	48
91203	Napthalene/Trimethylbicyclo[2.2.1]heptane	0.3	30.3	1.4	2.5	4.27	47
109068	Methylpyridine	0.3	210	1.4	5.81	29.18	49
108485	Diemethyl pyridine/Aldrich	0.3	48.2	1.4	2.56	6.43	49
100710	Ethylpyridine	0.3	69.4	1.4	3	9.52	49
78784	2-Methylbutane	0.3	3620	1.4	88.04	507.51	44
109660	Pentane	0.3	198	1.4	7.73	28.59	45
108087	2,4-Dimethylpentane	0.3	50	1.4	2.61	6.82	49
963772	Methyl cyclopentane	0.3	22	1.4	2.4	3.75	48
561764	2-Methylhexane	0.3	35.3	1.4	2.71	5.44	48
565693	2,3-Dimethylpentane	0.3	98	1.4	3.57	13.51	48
589344	3-Methylhexane	0.3	2300	1.4	49.02	321.63	46
108872	Methylcyclohexane	0.3	38	1.4	2.42	5.15	48
540841	2,2,4-Trimethylpentane	0.3	17	1.4	1.95	2.29	49
	C3 Hydrocarbon	0.3	51.4	1.4	5.57	9.7	42
	C4 Hydrocarbon	0.3	137	1.4	6.6	21.61	44
	C5 Hydrocarbon	0.3	6780	1.65	145.24	947.97	28
	C6 Hydrocarbon	0.3	294	1.6	18.06	51.83	30
	C7 Hydrocarbon	0.3	2390	1.5	56.64	333.96	38
	C8 Hydrocarbon	0.3	1420	1.5	39.44	199.78	36
	C9 Hydrocarbon	0.3	761	1.41	19.78	106.43	42
	C10 Hydrocarbon	0.3	191	1.5	11.75	30.7	37
	C11 Hydrocarbon	0.3	53.6	1.4	3.92	9.51	46
	C12 Hydrocarbon	0.3	395	5	23.81	59.1	22
	C13 Hydrocarbon	0.3	231	1.57	10.88	35.03	40
76641	Acetone	0.3	20.7	1.4	2.04	2.81	49
74986	Propane (ppmv)	1	62.9	1.4	2.97	8.65	48
106978	Butane (ppmv)	1	69	1.4	2.95	9.45	48
74840	Ethane (ppmv)	1	34.6	1.4	2.24	4.66	49
75285	Isobutane	0.3	34	1.5	3.95	6.38	38
79925	Camphene	0.3	5.2	1.4	1.65	0.81	49
592574	Cyclohexadiene	0.3	7.1	1.4	1.76	1.1	49
103651	Propynylbenzene	0.3	7.2	1.4	1.74	1.13	49
226666	Diethyl trisulfide	0.3	8.23	1.41	2.14	1.62	43
513359	Methylbutane	0.3	16	1.4	1.93	2.16	49
2511957	Dimethylcyclopropane	0.3	29	1.4	2.19	3.91	49
75832	Dimethylbutane	0.3	15	1.4	1.91	2.03	49
107835	Methylpentane/Isohexane	0.3	199	1.4	6.1	27.79	48
110543	Hexane	0.3	35	1.4	2.46	4.81	48
138863	Limonene	0.3	12.9	1.4	2.14	2.15	47
	Dimethylpentatnone	0.3	42.8	1.4	2.47	5.82	49
	Bromohexene	0.3	5.2	1.4	1.69	0.84	49
3728550	Ethylmethylcyclohexane	0.3	6.1	1.4	1.82	1.15	48
4316658	Trimethylhexene	0.3	11.9	1.4	1.85	1.64	49
1072168	Dimethyloctane	0.3	20.4	1.4	2.02	2.74	49
7785708	1-R-alpha-pinene/2-Pinene/2.6.6Trimethylbichyclo[3.3.1]hept-2-ene	0.3	29	1.4	2.18	3.91	49
496117	Indane	0.3	15.2	1.4	1.9	2.06	49
590738	2,2,-Dimethylhexane	0.3	168	1.4	4.97	23.3	49
251412	Thieno[3,2] thiopene	0.3	56.5	1.5	5.18	10.7	43
78853	Methacrolein	0.3	1710	1.4	35.74	239.18	49
106467	1,4-Dichlorobenzene	0.2	4.43	0.3	0.55	0.66	45
591764	Methylhexane	0.3	25	1.4	2.11	3.66	46
75694	Trichlorofluoromethane	0.3	5.2	1.4	1.64	0.79	49
75456	Difluorochloromethane	0.3	45	1.4	2.51	6.12	48
137631	Tetramethylcyclopentane	0.3	9.24	1.4	1.79	1.33	49
4926787	Ethylmethylcyclohexane	0.3	5.68	1.4	1.78	1.04	48
6069983	Methylmethylethylcyclohexane	0.3	6.17	1.4	1.73	1.01	49
543599	Chloropentane	0.3	5.2	1.4	1.65	0.8	49
592574	1,3-Cyclohexadiene	0.3	5.8	1.4	1.73	0.98	49
60779240	Methyl n-butyl disulfide	0.3	15.5	1.4	1.92	2.1	49
72437640	Propyl n-butyl disulfide	0.3	14.6	1.4	1.9	1.98	49
629196	Dipropyl disulfide	0.3	23.1	1.4	2.07	3.11	49

Methane was present in concentrations above laboratory detection limits in 98% of the sampling events where methane was sampled. Most of the areas investigated had atmospheric methane concentrations considerably higher than urban background concentrations (1.8–2.0 ppm_v) (National Oceanic and Atmospheric Administration [NOAA], 2006).

Benzene was identified as present in 38 of 50 sites sampled (76%). Many of the other chemicals had a high number of nondetects: many of the high maximums corresponded to a single sampling event where a natural gas well was experiencing a pressure malfunction, causing an uncontrolled emission event to the atmosphere. Trimethylbenzene specifically had several high-concentration observations recorded; however, the maximum value corresponded to the same ambient air sampling event with the pressure malfunction.

Table 3 compares maximum, median, and mean values for this study with those for three other studies from Colorado (McKenzie et al., 2012; Petron et al., 2012) and the U.K. (Hopkins et al., 2005). For methane, values from this study were higher than for the Petron study, which reported median methane levels from 1.81 to 1.89 ppm, which are typical background levels. For the alkanes ethane, propane, butane, pentane, and hexane, mean values from this study are generally comparable to values from the three other studies. For the other compounds, which include a number of aromatics (benzene, ethylbenzene, toluene, xylenes, trimethylbenzenes), maximum, median, and mean values from this study were generally higher than for the other studies. The right-hand-most column in Table 3 gives ambient average values for six of the compounds for California (because data were not available for Texas), to provide a rough idea of what background concentrations may be. The maximum values measured in this study are higher than the California ambient average values, except for methylene chloride. The mean values measured in this study are higher than the California ambient average values for three of the six compounds. Median values for this study were lower than the California values for four of the six compounds, likely due to the large number of nondetects in this study.

Table 3. Comparison with other studies

	This Study			Garfield County, Colorado^a		Birmingham, UK^b	Weld County, Colorado^c	State of California^d
Chemical	Max (ppbv)	Median (ppbv)	Mean (ppbv)	Max (ppbv)	Median (ppbv)	Range of Means (ppbv)	Range of Medians (ppbv)	Ambient Average (ppbv)
Methane (ppmv)	457	2.7	11.99				1.81–1.89
Benzene	592	0.89	18.53	4.39	0.30	0.25–0.7	0.02–0.1	4.6
Chloroform	2.58	0.3	0.45					0.006–0.13
Dichloromethane/Methylene chloride	1	0.3	0.34					1.1–2.4
Ethylbenzene	113	0.53	4.42	1.87	0.04
Styrene	43.4	0.37	1.91	0.80	0.035			10
Tetrachloroethene (PCE)	2.43	0.3	0.33					0.71
Toluene/Methylbenzene	276	2.55	19.45	21.0	0.48	0.7–1.9
Trichloroethene (TCE)	60.9	0.3	1.58					0.22
1,3,5-Trimethylbenzene	9.95	0.59	1.43	0.25	0.024
1,2,4-Trimethylbenzene	60.4	0.4	3.45	0.63	0.037
m- and p-Xylene	221	1.68	15.69	2.28	0.20
o-Xylene	39.4	0.85	3.19	0.83	0.05
Propylbenzene	23.5	1.4	2.08	0.14	0.02
Pentane	198	1.4	7.73	21.1	3.09	0.2–0.5	0.01–0.48
Methyl cyclohexane	38	1.4	2.42	5.98	0.92
Propane (ppmv)	62.9	1.4	2.97			0.8–2.8	0.1–3.0
Butane (ppmv)	69	1.4	2.95			0.9–2.8	0.04–1.24
Ethane (ppmv)	34.6	1.4	2.24			2.2–6.3
Isobutane	34	1.5	3.95			0.8
Methylpentane/Isohexane	199	1.4	6.1			0.15–1.1
Hexane	35	1.4	2.46	7.11	1.14	0.1–0.2
Notes: Values are reported by California Air Resources Board for California, except for benzene, which was reported for the South Coast Air Basin of California (Los Angeles metropolitan area).^a McKenzie et al. (2012). ^bHopkins et al. (2005). ^c Petron et al. (2012). ^dSeinfeld and Pandis (1998).

Correlations among concentrations of chemicals in ambient air samples

Further statistical analysis was performed using a Pearson's product-moment correlation coefficient to determine the strength of relationship between nonmethane volatile organic compounds (NMVOCs) in ambient air samples. To minimize the impact of missing measurements, compounds detected in fewer than 50% of the sites were excluded.

The correlation matrix for nonmethane compounds present at detectable levels in at least 50% of the sampling events identifies a relationship among various chemicals present and several strong correlations, as shown in Table 4. Notable correlation coefficients include those between benzene and toluene at r = 0.89, and between benzene to m- and p-xylene at r = 0.86. Toluene (methylbenzene) was also highly correlated to m- and p-xylene (1,3-dimethylbenzene and 1,4-dimethylbenzene) with r = 0.95. The BTEX compounds (benzene, toluene, ethylbenzene, and xylene) are often found together in petroleum derivatives, such as natural gas and gasoline.

Table 4. Pearson's correlation nonmethane volatile organic compounds

	Benzene	Chloromethane	Dichlorodifluoromethane	Toluene	m- and p- Xylene
Benzene	1.0
Chloromethane	0.0723	1.0
Dichlorodifluoromethane	−0.0791	−0.0999	1.0
Toluene	0.8927	0.1292	−0.1324	1.0
m- and p-Xylene	0.8658	0.0686	−0.1368	0.9543	1.0
C12 hydrocarbon	0.1823	−0.1746	0.2925	0.1744	0.257

Concentrations of numerous chemical constituents were significantly correlated to those of methane, as shown in Table 5. Given the sample size (n = 50), any correlation exceeding 0.2788 was significant at α = 0.05. 3-Methylhexane, the constituent with the strongest correlation to methane, is confirmed to be a constituent of natural gas condensate, according to a permit application submitted to the Texas Commission on Environmental Quality for a site in the Barnett Shale (XTO Energy, 2010). Chemicals significantly correlated to the presence of methane and identified as HAPs by EPA included hexachlorobutadiene, tetrachloroethene, 1,2,4-trichlorobenzene, and chloroform.

Table 5. Correlation of methane to chemical constituents

Chemical	Correlated to Methane
Methane	1.0
3-Methylhexane	0.81
C5 Hydrocarbon	0.68
C9 Hydrocarbon	0.68
2-Methylbutane	0.64
C7 Hydrocarbon	0.55
C8 Hydrocarbon	0.55
Trichlorofluoromethane (F-11)	0.53
Hexachlorobutadiene	0.36
Tetrachloroethene (PCE)	0.35
C4 hydrocarbon	0.34
Methylhexane	0.33
C6 Hydrocarbon	0.33
1,2,4-Trichlorobenzene	0.33
Diethyl trisulfide	0.29
Chloroform	0.29

Principal components analysis—Chemical constituents

To further examine the relationships among the chemical constituents, a principal components analysis (PCA) was performed with the goal of determining whether potential signatures could be identified for gas well emissions, or emissions associated with particular well operations. Data on 36 chemicals that were measured with detectable concentrations in at least 10% of samples were included in the PCA. However, this means that a compound may not have been detected in 90% of the samples. PCA assumes normal distributions, which may not hold with a large number of nondetect values. It is thus recommended, as mentioned later, that follow-up study be conducted with a larger sample size.

Table 6 shows eigenvalues determined for the chemical constituent data. According to Table 6, the first 3 PCA factors express almost half (49%) of the variance of these data, with the first 10 expressing 80% of the variance.

Table 6. Eigenvalues

	Eigenvalue	% total variance	Cumulative eigenvalue %	Cumulative %
1	8.13	22.57	8.1	22.57
2	6.12	16.99	14.24	39.56
3	3.27	9.07	17.51	48.63
4	2.69	7.47	20.20	56.11
5	1.98	5.49	22.17	61.60
6	1.77	4.92	23.94	66.51
7	1.53	4.26	25.48	70.77
8	1.27	3.52	26.75	74.29
9	1.18	3.28	27.93	77.57
10	1.06	2.94	28.98	80.51

Further analysis focused on the first two factors. Tables 7 and 8 indicate chemicals that contributed strongly to PCA factors 1 and 2, respectively (coefficients >0.7 in absolute value).

Table 7. Chemicals associated with factor 1

Chemical	Coefficient
o-Xylene	−0.90
Ethylbenzene	−0.87
1,2,4-Trimethylbenzene	−0.82
m- and p-Xylene	−0.81
1,3,5-Trimethylbenzene	−0.80
Toluene	−0.76
Benzene	−0.74

Table 8. Chemicals associated with factor 2

Chemical	Coefficient
Trichlorofluoromethane	0.75
Methane	0.72
C5 hydrocarbons	0.72
C8 hydrocarbons	0.71

Principal components analysis—Sample site scores

Further PCA evaluation was performed analyzing the scores of sample sites for the first two PCA factors. Figure 2 shows sample site scores plotted in relation to factors 1 and 2. We examined the distributions of sample site scores for their associations with aspects of well site operation. Nineteen sites had scores less than −0.5 on factor 1. These sites had up to 12 compressors per site. The 31 sites with scores higher than −0.5 on factor 1 had fewer compressors. Six of these 31 sites had nine compressors, and the remainder had two compressors per site. One site stands out by virtue of its high score (almost 12) for factor 2. This is the same site previously noted for maximum values for a number of chemicals, due to the well experiencing a pressure malfunction causing an uncontrolled emission event to the atmosphere.

Figure 2. PCA sample site scores factor 1:factor 2.

The PCA suggested that a multivariate signature of nearby gas well compression operations may be present in patterns of chemical concentrations found in ambient air samples (appearing as patterns associated with factor 1), and suggested that blown wells or other operations venting natural gas might also have a distinct signature (patterns associated with factor 2). An analysis of variance (ANOVA) F test was performed to test the null hypothesis that sample site scores for PCA factor 1 and factor 2 are independent of particular phases and equipment associated with nearby gas well operations. Based on the equipment observed at the nearest gas well to each sampling site, these sites were classified for five processes: (1) flaring versus not flaring; (2) hydraulic fracturing versus not hydraulic fracturing; (3) wells present versus absent; (4) compressors present versus absent; and (5) tanks or tank batteries present versus absent. The ANOVA tests results, conducted with a significance level of α = 0.05, are given in Table 9.

Table 9. ANOVA tests to determine whether factors 1 and 2 are independent of gas well operation phases and equipment

ANOVA Test	Accept Null Hypothesis for Factor 1?	Accept Null Hypothesis for Factor 2?
Flaring	Yes	Yes
Fracking	Yes	Yes
Wells	Yes	Yes
Compressors	No	Yes
Tanks	Yes	Yes

As shown in Table 9, the null hypothesis was accepted in all cases except in the case of compressors for factor 1. There is a significant difference in factor 1 in relation to compression, with more negative scores for factor 1 characterizing sites on which compressors are present. Operation of compressors is identified by a signature or fingerprint of chemicals in air masses known to be products of combustion and which contribute to differences among sites along factor 1 of the PCA (Table 7). Moreover, there was a quantitative relationship between number of compressors and scores for factor 1 (linear regression, two-tailed t test, P < 0.05), although it explained only 9% of the variation in factor 1 scores. Although the association of factor 2 with the well with the pressure malfunction suggests that there might also be a signature for well operations that vent large amounts of raw natural gas, ANOVA did not detect a statistically significant association of factor 2 scores with any of the well operations characterized here. It is possible that a larger sample size, with temporally repeated monitoring, would be necessary to detect emission signatures resulting from the episodic phases of well operation that risk large emissions of raw natural gas.

Study limitations and recommendations for future work

Limitations of this study include the small sample size and the fact that a positive correlation of contaminants with a common source does not rule out other sources. As mentioned above, PCA assumes normal distributions, which may not hold with a large number of nondetect values. It is thus recommended that follow-up study be conducted with a larger sample size.

Sampling was performed in primarily residential areas with few potential emission sources unrelated to natural gas extraction and production, other than light traffic; however, it is possible that some transportation emissions or other combustion sources may have contributed somewhat to some chemical concentrations recorded. Future studies may evaluate to what degree traffic emissions contribute to overall atmospheric conditions in areas with and without natural gas emission sources. Future work should include collecting 1-hr samples along with field meteorological data, to determine whether the well sites are upwind, downwind, or crosswind during the sampling period. This would indicate whether the drilling/production sites are likely sources of the pollutants. In addition, carbon isotope analysis, which can distinguish among various sources of organic compounds, should be used to provide evidence that the methane is from natural gas sites, rather than wetlands, landfills, wastewater treatment plants, or other potential sources.

The assumption should not be made that all airborne chemicals can be found in the tests performed. In fact, numerous chemicals may exist that are not properly identified through the methodology and laboratory analysis performed in this study, as it is specific to chemicals identified on the target list of analytes (TO-14A), light hydrocarbons and tentatively identified compounds.

Summary

This exploratory study found methane and 101 other chemicals to be present in the atmosphere in and around shale gas well sites located in residential areas in the DFW Metroplex. Methane was present in concentrations above laboratory detection limits in 49 out of 50 sampling events. Most of the areas investigated had atmospheric methane concentrations considerably higher than reported urban background concentrations (1.8–2.0 ppm_v). Other chemical constituents were found to be correlated with presence of methane.

A principal components analysis identified multivariate patterns of concentrations that potentially constitute signatures of emissions from different phases of operation at natural gas sites. The first factor identified through the PCA proved most informative. Extreme negative values were strongly and statistically associated with the presence of compressors at sample sites. Aromatics in particular contributed to the factor associated with compressor stations. An extreme positive value on factor 2 suggest that there may also be a signature of operations that release large amounts of natural gas, but this singular association was not statistically significant in these data.

A follow-up study with a larger sample size and more rigorous analysis is recommended.